mfakto: an OpenCL program for Mersenne prefactoring

[kracker]

Originally Posted by Rodrigo

The exponents were in the 67M range (TF).

My 7770 is as it came, no adjustments made to it. The only tweak I've made to mfakto 0.13 is to change the VectorSize value from the default 4 down to 2, as suggested by the program itself the first time I ran it.

Have you made any other adjustments to increase the output?

Rodrigo

I pushed GPUSievePrimes around 10k up from default, GPUSieveProcessSize to 24 from 16.

[LaurV]

Let's start by telling you that I have to go back 36 pages and read this topic from the start. I got my hands on a XFX HD 7970 and playing with it since last Friday. I may keep it for a while, but after running it till now, I am still remaining definitively a nVidia guy (sorry to say! reasons are multiple, not only related to performance or consumption, but goes also into scrypt mining, etc. not related to the current subject, we may discuss more in a different topic if someone is interested).

But my problem now is not with the card, but more with mfakto. Using default .ini file the performance was lousy, and the program advised me to change whatever variable from 4 to 2 (kudos to Bdot for not letting beginners like me scratch their head in vain!), which I did (something about magical-disappearing registers). With new settings, I can get about 390GHzD/D out of it with P95 running (GPU sieving) and about 420 with P95 stopped (GPU sieving too, there is no question about keeping my CPU busy with sieving, I would not try it if that was the case!). This for factoring in the 332M range of the expos (I took a bunch of them to 74 bits, founding 6 factors on the way, as one could check with the DB, in the last 5 days).

Everything wonderful up to here. Hip-hip hoorays go to Bdot.

But then my freaking fingers tickled and I run --perftest, and start playing with number of primes to sieve (default ~80k in the ini file). For any other values except the default one, the -sp test fails (always missing 2-3 factors from all the bunch). I tried numbers around 60k, 100k, 200k, and larger, close to the max 1M. It seems that the card would get better results for smaller value there, but what the hack, it always missing factors. I am thinking that the number I use there should be a multiple of something or have special properties? Otherwise I can't find an explanation. I tried also playing with the clock (like in "reducing it") and the card consumption (like "increasing it", give it more voltage), but the results are the same. In this status, my money are either on (1) some properties of the number which I fail to acknowledge (and therefore I am giving a bad parameter to the program, but I would wonder if Bdot missed to warn the user when rubbish is put in, I saw Bdot is very diligent with warning messages!) or either on a (2) software bug.

Therefore the current post... Till a reply, I will go back to read the topic from the beginning, to see what I missed.

Any idea?

[kracker]

Originally Posted by LaurV

~~~~

Hmm, I don't think I've tried -st on other SievePrime than the default.

By the way, what catalyst version are you using? it shouldn't drop to 390 from 420... Just curious what GHz do you get in the LL or DC range?

[LaurV]

Didn't try current LL range yet. Queued some work from Chris for the next days.

OTOH, I always edit my posts in the next 10-20 minutes after they were posted, to repair grammar, typos, accentuate ideas, etc. It is not a good habit of you to include my last posts into reply, so everybody can see my stupidity before I am fast enough to mask it . Please delete it! (anyhow, it is not good to include a quote to a full post, especially if you reply to the last post, that's what the netiquette says, beside of the fact that it takes futile space onthe servers).

[LaurV]

Originally Posted by Bdot

Third, the AMD drivers. Sadly, the latest windows drivers (13.4) make mfakto consume almost one CPU core, even when GPU sieving. When other programs (like prime95) use a lot of CPU, then mfakto's high CPU load goes away, but at the cost of some 10-20% throughput.

That is the first "Bingo!" related to my problem. Thanks. I am glad I started to read the topic viceversa! (from the newest posts to the oldest). The discussion is not necessarily connected, but it can provide a faster solution, hehe.

What drivers are recommended? (Win7 64 bits)

Still reading.

[kracker]

Originally Posted by LaurV

That is the first "Bingo!" related to my problem. Thanks. I am glad I started to read the topic viceversa! (from the newest posts to the oldest). The discussion is not necessarily connected, but it can provide a faster solution, hehe.

What drivers are recommended? (Win7 64 bits)

Still reading.

13.1. They aren't available on AMD's site because they are older, so techpowerup has them here. I also recommend using AMD's complete cleanup tool here to remove the driver completely.

Oh and sorry for the full quote, I really wasn't thinking..

[Rodrigo]

Originally Posted by kracker

Is this in the LL range ~65M? On my 7770 non OC'ed I am getting around 160 Ghz/days.

kracker, by any chance were you running Prime95 concurrently when you got this output?

The reason I ask has to do with Bdot's observation (just recently quoted by LaurV) that:

Sadly, the latest windows drivers (13.4) make mfakto consume almost one CPU core, even when GPU sieving. When other programs (like prime95) use a lot of CPU, then mfakto's high CPU load goes away, but at the cost of some 10-20% throughput.

What I'm thinking is that if you take 10% from 160 GHz-days/day, you get 144 which is very close to what I'm getting with no tweaks beyond the VectorSize value, so could it be that it's not necessary to make further manual adjustments? (I was running Prime95 when I ran my mfakto tests.)

Just to confuse things further , FWIW I'm on "Catalyst Version" 12.10 according to AMD's Catalyst Control Center but Windows Device Manager says that the "driver version" is 9.2.0.0, dated 9/27/2012. I'm never sure which number is the "real" one, but if I had to bet on one I'd put my money on 12.10.

Rodrigo

[kracker]

Originally Posted by Rodrigo

kracker, by any chance were you running Prime95 concurrently when you got this output?

Nope.

Just to confuse things further , FWIW I'm on "Catalyst Version" 12.10 according to AMD's Catalyst Control Center but Windows Device Manager says that the "driver version" is 9.2.0.0, dated 9/27/2012. I'm never sure which number is the "real" one, but if I had to bet on one I'd put my money on 12.10.

Rodrigo

the driver version is, the 9.2.0.0. the package version is 12.10, that's what everyone means when they say driver version. 12.10 is not affected by the cpu bug.

[Rodrigo]

Originally Posted by kracker

the driver version is, the 9.2.0.0. the package version is 12.10, that's what everyone means when they say driver version. 12.10 is not affected by the cpu bug.

All right, I'll stay on 12.10 'til things get sorted out. Thanks.

Rodrigo

[Rodrigo]

More observations on version 0.13:

Often (not all the time), I've noticed a distinct lag when repositioning windows, opening websites, or even moving the mouse cursor around the screen -- especially just after sending a file to the printer. The mouse cursor lag is particularly pronounced as it seems to just sit there after I move the mouse, and then the cursor pops up way off somewhere else, making it difficult to select or click on things on the screen.

None of this happens with version 0.12.

The shine is coming off a bit from 0.13. As this is a production machine, I'm going back to 0.12 for the time being, even though Prime95 takes a hit and the TF yield is lower even with three mfakto instances running.

Anybody else getting this lag? If I can get rid of it, it'll make 0.13 viable again for me.

Rodrigo

[Jayder]

Did you try lowering GPUSieveSize in mfakto.ini? That was what made the big difference for me.