The latest version of Genefer, v. 3.2.2, has been released and is live for both short and World Record tasks. Some notable changes:

1) The OpenCl and CPU apps are significantly faster than the previous versions.

2) It is likely that the OpenCL app is now faster than the CUDA app on any Nvidia GPU, but especially on newer (Kepler or later, 6xx or newer) GPUs. On the app selection page, the OpenCL app now appears before the CUDA app. NOTE: On Nvidia cards, the OpenCL driver hogs a CPU core. If you have an Nvidia GPU, you may chose between the faster app (OpenCL) or that app that uses very little CPU (CUDA).

3) The CPU app now supports the FMA3 instructions on Intel Haswell CPUs, which makes them even faster.

4) The bug in the Mac ATI app has been fixed.

____________
My lucky number is 75898⁵²⁴²⁸⁸+1

Just tried, and I received Genefer 3.0.1 ?

Or : is it the name of the new app ?

R280X, slow C2D CPU and some benchmarks using different version of Genefer.
This host is on PRPNet so I should know in an hour about 827258^524288+1 test. I'm currious not only about time but also Err value.
EDIT: 827258^524288+1 is a probable composite. (RES=6dfe04ee1cce65c9) (3102549 digits) (err = 0.3906) (time = 1:19:20)

Note Err value and also note Genefer Mark.

Version 3.1.2.7
658332^524288+1 Time: 824 us/mul. Err: 0.5000 3050541 digits
360204^4194304+1 Time: 5.38 ms/mul. Err: 0.5000 23305854 digits
Genefer Mark = 74.

Version 3.2.0.0
658332^524288+1 Time: 551 us/mul. Err: 0.2266 3050541 digits
360204^4194304+1 Time: 3.91 ms/mul. Err: 0.1953 23305854 digits
Genefer Mark = 102.

Version 3.2.2.0
Generalized Fermat Number Bench
658332^524288+1 Time: 426 us/mul. Err: 0.1875 3050541 digits
360204^4194304+1 Time: 3.42 ms/mul. Err: 0.1719 23305854 digits
Genefer Mark = 116.
____________
My stats

I started BOINC with a GFN short and copied the Genefer.exe from a slot directory.
I get the following when I try to run from the command line:

I started BOINC with a GFN short and copied the Genefer.exe from a slot directory.
I get the following when I try to run from the command line:

C:\Users\Downloads\genefer-3.2.2>primegrid_genefer_3_2_2_0_3.01_windows_intelx86__atiGFN.exe -l This version of C:\Users\KarpinFamily\Downloads\genefer-3.2.2\primegrid_genefer_ 3_2_2_0_3.01_windows_intelx86__atiGFN.exe is not compatible with the version of Windows you're running. Check your computer's system information to see whether you need a x86 (32-bit) or x64 (64-bit) version of the program, and then contact the software publisher.

I am running Windows 7 64 bit
Is this application 32 bit? How do I run it from the command line in 32 bit mode? Where can I download the 64 bit version?

When I try to run the exe from the command line I also get a comment box titled "Unsupported 16-Bit Application". Doubt that.

As you can see there has been a significant improvement in the OpenCL code (and also in the CPU version for some CPUs). The vast majority of the credit for this goes to Yves and his rediscovery and implementation of the Z-transform algorithm.

Cheers

- Iain

Version 3.1.2.7
658332^524288+1 Time: 824 us/mul. Err: 0.5000 3050541 digits
360204^4194304+1 Time: 5.38 ms/mul. Err: 0.5000 23305854 digits
Genefer Mark = 74.

Version 3.2.0.0
658332^524288+1 Time: 551 us/mul. Err: 0.2266 3050541 digits
360204^4194304+1 Time: 3.91 ms/mul. Err: 0.1953 23305854 digits
Genefer Mark = 102.

Version 3.2.2.0
Generalized Fermat Number Bench
658332^524288+1 Time: 426 us/mul. Err: 0.1875 3050541 digits
360204^4194304+1 Time: 3.42 ms/mul. Err: 0.1719 23305854 digits
Genefer Mark = 116.

I'm currious not only about time but also Err value.

We've improved the error handling a bit in most of the 3.2.2 apps and released 3.2.3 as BOINC 3.03.

Note that the 3.2.2/3.01 Windows CUDA and Windows CPU apps were already working as intended and have not been updated.
____________
My lucky number is 75898⁵²⁴²⁸⁸+1

One way to soften the impact of the "cpu hog" effect is to turn hyperthreading ON. GeneferOCL still wants 100% of a "core", but you've got twice as many "cores" to play with. There's no apparent impact on the overall runtime of GFN-short units, at least... the GPU is the deciding factor there; I've run a couple as a test using the newest stock app. This is on a 2600K, Ubuntu 14.04, GTX770.

Of course the down-side is that if you're running LLR on the other cores, they will take roughly twice as long to run (depending on how many you've got going). But if you're cpu-sieving, you should have HT ON anyway.

G

I don't know how advanced some of the volunteer programmers are here at Primegrid (thank you all) but, I came across some very advanced optimization info and testing software from a guy named Agner Fog that 'maybe' of help for continuing to make GFN app (and other Primegrid apps) better, faster, more efficient. Hopefully there's something helpful there.

http://agner.org/optimize/

____________
Largest Primes to Date:
As Double Checker: SR5 109208*5^1816285+1 Dgts-1,269,534
As Initial Finder: SR5 243944*5^1258576-1 Dgts-879,713

Yes it does. If it says it's using AVX, it's using AVX.

Please let us know about any irregularities.
____________
My lucky number is 75898⁵²⁴²⁸⁸+1

Yes it does. If it says it's using AVX, it's using AVX.

Please let us know about any irregularities.

It seems to be running just fine, but shouldn't the program have chosen FMA instead of AVX or am I just confused? I don't have any benchmarks to go off of since its been a while since I've run an AVX CPU Genefer app.

We discovered that the CPU version of Genefer was trying to run the FMA3 transform on AMD Bulldozer CPUs -- which have FMA4 instructions but not FMA3.

That's been fixed in version 3.2.4 (BOINC v. 3.04), which is now live. Only the CPU Genefer apps have changed.

Please let me know if any unexpected happens.
____________
My lucky number is 75898⁵²⁴²⁸⁸+1

Is there a difference between the version 3.01 OpenCL NVidia Genefer WR program and the version 3.03 OpenCL Genefer WR? I noticed when I downloaded the workunit on the 28th of July it was 3.01, since then the program itself has been updated to 3.03. If I understand correctly the underlying version is 3.2.2 and that hasn't changed. Is that just some minor improvements and adjustments or something more?

Also, thanks for all the hard work improving these programs. Theres definitely a big speedup in the processing... from 259 hours to 216 hours. That's a huge improvement!

primegrid_genefer_3_2_3_0_3.03_x86_64-pc-linux-gnu__OCLcudaGFN downloaded for me fine and appears to be running OK on my Ubuntu 14.04 (64-bit) box, with driver 331.79, GTX770. ETC per stderr.txt is 7:32:35, which is basically the same as several verified results I reported a few weeks ago with the prior version.

--Gary
EDIT: obviously this comment applies to the GPU app.

primegrid_genefer_3_2_3_0_3.03_x86_64-pc-linux-gnu__OCLcudaGFN is also working with Linux Mint LMDE (another Debian derivative) using 64-bit Nvidia drivers from the distro's repo.

One result validated, and one failed with maxERR exceeded. I haven't been watching the GPU internal temperature lately; it's now at 71 C, previously 67 C. Cleaning the dust off the case grills didn't help, but manually setting the GPU fan speed to 100% drops the GPU temperature to 61 C. Looks like I can't trust the factory default GPU fan speed setting to keep the system cool enough during GFN.

Would it be possible, when a WU is sent out, IF there is a CPU asking for work, to send one duplicate to a GPU and the other to a CPU? i.e., never send the initial two both to CPU, and then create the potential for a 2-month wait before either task even gets re-sent. Just thinking aloud, although I really would like to test a WR task on my CPU. I think I've asked before, but please put me at the top of the list for a trial of GFN-WR on the CPU.

What was the last important change you implemented to PG which was "easy"?

Yes, I am half-joking, my main reason for posting earlier was to say "Wow, the CPU Generfer app is fast, it used to be basically a GPU-or-almost-nothing thing.".

But now it isn't, so I was just looking ahead.