Testing is complete, the new app is live!

New features:

-AMD Navi GPU supported
-CPU multi-threading supported (up to 64 threads) using PrimeGrid preferences web page settings like LLR
-AVX512dq supported
-GPU app should be about 5% faster overall

Improvements came from the upcoming (hopefully?) Wieferich/WallSunSun search apps i've been working on.

Testing is complete, the new app is live!

New features:

-AMD Navi GPU supported
-CPU multi-threading supported (up to 64 threads) using PrimeGrid preferences web page settings like LLR
-AVX512dq supported
-GPU app should be about 5% faster overall

Improvements came from the upcoming (hopefully?) Wieferich/WallSunSun search apps i've been working on.

The speedup is nice to see, but my machine (273226) is dang near unusable when it's running. Sure, it was a little choppy before when doing anything graphical (videos, transitions), but now it's a slideshow just moving my mouse across the empty desktop. When I first encountered it, I thought the system had crashed.

Is there any way to help mitigate this aside from the no bueno "suspend when active" setting? Something like the Bn settings in other apps? (just updated the drivers to latest and now the mouse is smooth, but that's it.)
____________
Eating more cheese on Thursdays.

Awesome thank you. I was one of the mad ones who has been doing it on CPU!

Does PG automatically download new app when it fetches work?

I am running with app_config because I have multi GPUs and I need the app_config to regulate which jobs are running on them. What is the call out for the new app for app_config?

I see it supports AVX-512. What about regular AVX and AVX2?
____________
Reno, NV

I see it supports AVX-512. What about regular AVX and AVX2?

AVX and AVX2 were always supported by AP27. The new app adds AVX512 support.

my machine (273226) is dang near unusable when it's running. Sure, it was a little choppy before when doing anything graphical (videos, transitions), but now it's a slideshow just moving my mouse across the empty desktop.

My PC crash/shut down with an Video Error, after a few seconds with an AP27 Search v2.08 (opencl_ati_AP27). ...again and again.

Ryzen 5 3600, 3x RX 5700 with Adrenalin 20.4.2. and Win10 Pro.

Sorry to interrupt on the problem-solving, but the new AP app with MT is fantastic! Estimated time on CPU with 4 threads is half that of on my integrated GPU! (8CU APU)
____________
SHSIDElectronicsGroup@outlook.com

waiting for a TdP prime...
Proth "SoB": 44243*2^440969+1

What the
https://www.primegrid.com/workunit.php?wuid=668683388

Such a rare sight! I got a _19 task!!!!!! AND successfully completed it!
____________
SHSIDElectronicsGroup@outlook.com

waiting for a TdP prime...
Proth "SoB": 44243*2^440969+1

What the
https://www.primegrid.com/workunit.php?wuid=668683388

Such a rare sight! I got a _19 task!!!!!! AND successfully completed it!

What the
https://www.primegrid.com/workunit.php?wuid=668683388

Such a rare sight! I got a _19 task!!!!!! AND successfully completed it!

This specific task failed on all GPUs. We're working on the fix.

You did it successfully on CPU.

Was there an algoritm problem or was this just like one of those "edge" cases where LLR's maxerr>0.4?

What the
https://www.primegrid.com/workunit.php?wuid=668683388

Such a rare sight! I got a _19 task!!!!!! AND successfully completed it!

This specific task failed on all GPUs. We're working on the fix.

You did it successfully on CPU.

The new app takes advantage of a NVIDIA feature that's been around for a few years called Compute Preemption. Using a Pascal or newer GPU under Windows 10 RS4 (April 2018), software developers no longer have to worry about kernel runtime being too long causing sluggish screen updates. The screen takes priority over the kernel. Before this, developers had to pick a small enough kernel work size for each different gpu encountered, and hope that screen refresh would be OK. That small work size can lead to inefficient utilization of modern GPUs with lots of compute units.

This is interesting. Actually explains why I've never had a stuttering problem with the old app on my Win 10 boxes which always made me curious (as some of them have hand me down GPUs from my Win 7 system and did stutter back then). Sad that it can't be fixed, though. At least this machine is (hopefully) in its last days of primary use if Intel can ship me the CPU for my new main rig.

This is interesting. Actually explains why I've never had a stuttering problem with the old app on my Win 10 boxes which always made me curious (as some of them have hand me down GPUs from my Win 7 system and did stutter back then). Sad that it can't be fixed, though. At least this machine is (hopefully) in its last days of primary use if Intel can ship me the CPU for my new main rig.

When fixing the bug that caused that workunit to fail, I also made a change to the app for better display response on AMD or Win7 systems.

This is interesting. Actually explains why I've never had a stuttering problem with the old app on my Win 10 boxes which always made me curious (as some of them have hand me down GPUs from my Win 7 system and did stutter back then). Sad that it can't be fixed, though. At least this machine is (hopefully) in its last days of primary use if Intel can ship me the CPU for my new main rig.

When fixing the bug that caused that workunit to fail, I also made a change to the app for better display response on AMD or Win7 systems.

But I have a question: PG says the app is v2.08, but the app says v3.0.
The problem has been existing before: PG v2.06, app v2.6
What's up with the naming?
____________
SHSIDElectronicsGroup@outlook.com

waiting for a TdP prime...
Proth "SoB": 44243*2^440969+1

But I have a question: PG says the app is v2.08, but the app says v3.0.
The problem has been existing before: PG v2.06, app v2.6
What's up with the naming?

An updated GPU app has been published. Version number has been bumped to 2.09 for all platforms (although CPU app is really unchanged).

- Fixed crash of GPU application on some specific tests/sequences;
- Screen responsibility should be improved (less lag) if you're using either AMD card either Windows 7/8. It has been not tested much on all possible combinations, so please report your feelings here.

Known issues:

- AMD drivers are buggy, often they cannot handle complex computing kernels like used this application. If something is not working on AMD card, you may post reports here but most probably it'll be a driver's problem. OpenCL driver is a very complex thing - really, it's a true compiler from C language to video chip machine code. Lot of bugs can be hidden there - driver may fail to compile syntactically correct source code, may compile it but generate incorrect video chip machine code, etc.

Why is Stderr output as to what's being tested suppressed in the new version?

I always like to see the detail for example :-


Stderr output
<core_client_version>7.14.2</core_client_version>
<![CDATA[
<stderr_txt>
AP26 OpenCL 10-shift search version 2.6 by Bryan Little and Iain Bethune
Compiled May 9 2019 with GCC 8.1.0
Command line: projects/www.primegrid.com/ap27_2.6_opencl_windows64.exe 46909826 46909946 1280
GPU Info:
Name: GeForce RTX 2080
Vendor: NVIDIA Corporation
Driver: 442.23
Using L2 cache for sieve.
K 46909826 done in 15 sec. Kernel queue: 8
K 46909827 done in 15 sec. Kernel queue: 7
K 46909828 done in 14 sec. Kernel queue: 7
K 46909829 done in 15 sec. Kernel queue: 8
K 46909830 done in 15 sec. Kernel queue: 8
K 46909831 done in 14 sec. Kernel queue: 7
K 46909832 done in 15 sec. Kernel queue: 7
K 46909833 done in 14 sec. Kernel queue: 8
K 46909834 done in 15 sec. Kernel queue: 7
K 46909835 done in 15 sec. Kernel queue: 7
K 46909836 done in 14 sec. Kernel queue: 7
K 46909837 done in 15 sec. Kernel queue: 8
K 46909839 done in 15 sec. Kernel queue: 8
K 46909840 done in 15 sec. Kernel queue: 8
K 46909841 done in 15 sec. Kernel queue: 7
K 46909842 done in 17 sec. Kernel queue: 6
K 46909843 done in 14 sec. Kernel queue: 9
K 46909844 done in 14 sec. Kernel queue: 8
K 46909845 done in 15 sec. Kernel queue: 8
K 46909846 done in 15 sec. Kernel queue: 8
K 46909847 done in 15 sec. Kernel queue: 8
K 46909848 done in 14 sec. Kernel queue: 8
K 46909850 done in 15 sec. Kernel queue: 8
K 46909852 done in 14 sec. Kernel queue: 9
K 46909853 done in 16 sec. Kernel queue: 6
K 46909855 done in 15 sec. Kernel queue: 9
K 46909856 done in 14 sec. Kernel queue: 7
K 46909857 done in 17 sec. Kernel queue: 6
K 46909859 done in 16 sec. Kernel queue: 6
K 46909860 done in 15 sec. Kernel queue: 7
K 46909862 done in 14 sec. Kernel queue: 8
K 46909864 done in 15 sec. Kernel queue: 7
K 46909865 done in 14 sec. Kernel queue: 8
K 46909866 done in 15 sec. Kernel queue: 7
K 46909867 done in 15 sec. Kernel queue: 7
K 46909868 done in 14 sec. Kernel queue: 8
K 46909869 done in 15 sec. Kernel queue: 7
K 46909870 done in 15 sec. Kernel queue: 7
K 46909871 done in 15 sec. Kernel queue: 9
K 46909872 done in 15 sec. Kernel queue: 7
K 46909873 done in 18 sec. Kernel queue: 7
K 46909874 done in 15 sec. Kernel queue: 7
K 46909875 done in 15 sec. Kernel queue: 8
K 46909877 done in 15 sec. Kernel queue: 6
K 46909879 done in 16 sec. Kernel queue: 7
K 46909880 done in 15 sec. Kernel queue: 7
K 46909881 done in 14 sec. Kernel queue: 7
K 46909883 done in 15 sec. Kernel queue: 7
K 46909884 done in 16 sec. Kernel queue: 7
K 46909885 done in 15 sec. Kernel queue: 9
K 46909886 done in 14 sec. Kernel queue: 8
K 46909887 done in 15 sec. Kernel queue: 8
K 46909888 done in 14 sec. Kernel queue: 8
K 46909889 done in 15 sec. Kernel queue: 6
K 46909890 done in 15 sec. Kernel queue: 7
K 46909891 done in 14 sec. Kernel queue: 7
K 46909892 done in 15 sec. Kernel queue: 8
K 46909893 done in 15 sec. Kernel queue: 8
K 46909894 done in 16 sec. Kernel queue: 8
K 46909896 done in 15 sec. Kernel queue: 8
K 46909898 done in 15 sec. Kernel queue: 8
K 46909899 done in 15 sec. Kernel queue: 7
K 46909900 done in 15 sec. Kernel queue: 7
K 46909902 done in 14 sec. Kernel queue: 7
K 46909903 done in 15 sec. Kernel queue: 7
K 46909905 done in 15 sec. Kernel queue: 7
K 46909906 done in 14 sec. Kernel queue: 8
K 46909908 done in 16 sec. Kernel queue: 7
K 46909909 done in 14 sec. Kernel queue: 7
K 46909910 done in 15 sec. Kernel queue: 8
K 46909911 done in 15 sec. Kernel queue: 7
K 46909912 done in 15 sec. Kernel queue: 7
K 46909914 done in 15 sec. Kernel queue: 8
K 46909915 done in 17 sec. Kernel queue: 8
K 46909916 done in 17 sec. Kernel queue: 6
K 46909917 done in 15 sec. Kernel queue: 8
K 46909918 done in 15 sec. Kernel queue: 7
K 46909919 done in 15 sec. Kernel queue: 7
K 46909920 done in 15 sec. Kernel queue: 7
K 46909921 done in 16 sec. Kernel queue: 6
K 46909922 done in 15 sec. Kernel queue: 7
K 46909923 done in 15 sec. Kernel queue: 8
K 46909924 done in 14 sec. Kernel queue: 9
K 46909925 done in 16 sec. Kernel queue: 8
K 46909926 done in 14 sec. Kernel queue: 8
K 46909927 done in 16 sec. Kernel queue: 7
K 46909928 done in 15 sec. Kernel queue: 7
K 46909930 done in 15 sec. Kernel queue: 5
K 46909931 done in 15 sec. Kernel queue: 6
K 46909933 done in 16 sec. Kernel queue: 7
K 46909934 done in 14 sec. Kernel queue: 10
K 46909935 done in 15 sec. Kernel queue: 7
K 46909937 done in 14 sec. Kernel queue: 8
K 46909938 done in 15 sec. Kernel queue: 8
K 46909939 done in 15 sec. Kernel queue: 6
K 46909940 done in 16 sec. Kernel queue: 7
K 46909941 done in 15 sec. Kernel queue: 9
K 46909942 done in 16 sec. Kernel queue: 6
K 46909943 done in 14 sec. Kernel queue: 9
K 46909946 done in 18 sec. Kernel queue: 6
Workunit complete.
00:22:14 (22992): called boinc_finish(0)

</stderr_txt>
]]>


Just curious, not complaining :)


____________
My lucky numbers 1059094^1048576+1 and 224584605939537911+81292139*23#*n for n=0..26

It's not needed.

It's not needed.