processed dc and tc posts

[ATH]

Originally Posted by Jan S

But something is wrong: I choose "double-check using PRP with proof"(manual assigment), all exponents are "Cat 0" and i have only 7 days to go. :-(

If you complete 0.1% of each of them, they will update to 30 days, because they need to be started within 7 days.

It is because of the unverified LL test, so technically it is not a PRP double check...but it should be fixed somehow if possible.

[Uncwilly]

List updated. There are 2 Cat 1 exponents available.

[ZacHFX]

MadPoo already has the Cat 1s.

I grabbed 63349189. I tried to get the other Cat 3's, but it wouldn't let me, as they have assigned TF.

I have also taken the 14 Cat 2's with a suspect + unverified result.

[Uncwilly]

Originally Posted by ZacHFX

I tried to get the other Cat 3's, but it wouldn't let me, as they have assigned TF.

That is likely MadPoo taking them to 75 or 76. They should be available soon.

[techn1ciaN]

Claimed:

PRP=1,2,332242501,-1

Curious on how quickly my 5700 XT can get through one of these.

Also — Kriesel seems to have already taken care of

PRP=1,2,100006129,-1

back in September.

[Uncwilly]

Originally Posted by techn1ciaN

Also — Kriesel seems to have already taken care of

I don't have a custom query to find those and check on their status. I should PM Aaron about that.

[kriesel]

Originally Posted by techn1ciaN

PRP=1,2,332242501,-1

Curious on how quickly my 5700 XT can get through one of these.

Rough estimate; 2 weeks on a Radeon VII, so ~6 weeks on a well behaved 5700XT and Gpuowl v6.11-357 to 380 for ~best PRP-only speed. (My 5700 XT is temperamental to say the least. About the time I'm ready to RMA it under warranty, it behaves for a week or two. Suspecting it prefers direct PCIe slot not extenders, and also cooler ambient temps. Its quirkiness can really balloon total elapsed time. It's currently unplugged but may get another try in cooler weather. Good luck with yours.)

[Jan S]

Claimed:

    PRP=1,2,332242501,-1

 Curious on how quickly my 5700 XT can get through one of these.

I worked on 369452123, it took 66 days on 5500XT.

[techn1ciaN]

Originally Posted by kriesel

Rough estimate; 2 weeks on a Radeon VII, so ~6 weeks on a well behaved 5700XT and Gpuowl v6.11-357 to 380 for ~best PRP-only speed.

How are you getting that number? I quickly bumped the line to the top of my work file to check and GPUOwl is giving an ETA of closer to 4 weeks (30d16h; 8.03 ms/iter). I'm actually using GPUOwl 7.2 and also running a modest GPU undervolt.

Mersenne.ca's benchmarks would seem to indicate that the 5700 XT should be about half as fast as the R7 for a given FFT length, give or take 100 µs/iter, meaning my performance is in the range of expectations leaning towards slightly worse. That could be the undervolt, the less-than-optimal GPUOwl version, GDDR6 starting to give up ground to HBM at 100M-digit FFT lengths, or some combination of all three.

[kriesel]

Originally Posted by techn1ciaN

How are you getting that number? I quickly bumped the line to the top of my work file to check and GPUOwl is giving an ETA of closer to 4 weeks (30d16h; 8.03 ms/iter). I'm actually using GPUOwl 7.2 and also running a modest GPU undervolt.

Mersenne.ca's benchmarks would seem to indicate that the 5700 XT should be about half as fast as the R7 for a given FFT length, give or take 100 µs/iter, meaning my performance is in the range of expectations leaning towards slightly worse.

Actual run times on Radeon VII 332M 13-14 days on Windows. Depending on fft length (~2:1 effect!), gpuowl version, and driver, etc it's capable of up to ~510 GHD/day in Linux with ROCm, ~480 on Windows. The mersenne.ca #s seem low for Radeon VII. Since the 5700XT is a newer model its benchmarks probably got run by faster gpuowl versions.

Various timings I've compared have indicated RX5700XT as ~30 - 45(rare)% of the Radeon VII's speed. Perhaps it reflects also upclocking Hynix Radeon VII HBM vram.
Haven't systematically benchmarked my 5700XT or tabulated past runs yet. It's not reliable enough for that.
Its timings are ~10-13ms/it on 480M. Should take about 2 months for that PRP. But nearly all of that remains, after 4+ months, because of EE or error code 31 or this sort of thing:

2021-06-25 16:51:29 test/rx5700xt 480216091 OK        0 loaded: blockSize 400, 0000000000000003
2021-06-25 16:51:29 test/rx5700xt validating proof residues for power 9
2021-06-25 16:51:29 test/rx5700xt Proof using power 9
2021-06-25 16:51:59 test/rx5700xt 480216091 OK      800   0.00%; 11234 us/it; ETA 62d 10:33; 5cecef28e9d008d6 (check 21.04s)
2021-06-25 17:29:39 test/rx5700xt 480216091 OK   200000   0.04%; 11293 us/it; ETA 62d 17:51; 50d5cdb29a52db16 (check 9.60s)
2021-06-25 18:43:24 test/rx5700xt GPU -> Host read #0 failed (check fb6f3e37 vs 1)
2021-06-25 18:43:24 test/rx5700xt GPU -> Host read #1 failed (check fb854b90 vs 3c72)
2021-06-25 18:43:24 test/rx5700xt GPU -> Host read #2 failed (check fb6f3e37 vs 1)
2021-06-25 18:43:33 test/rx5700xt Exiting because "GPU -> Host persistent read errors"

...

2021-10-14 13:35:07 test/rx5700xt 480216091 OK 61100000  12.72%; 10964 us/it; ETA 53d 04:28; e0e728177057a0b8 (check 5.50s)
2021-10-14 13:53:29 test/rx5700xt 480216091 OK 61200000  12.74%; 10965 us/it; ETA 53d 04:16; 2ec90b1b815bf1e4 (check 5.52s)

It has had these issues on at least 3 different systems tried.

What OS are you running?

If Windows, you may be paying a considerable penalty by running gpuowl v7.2 instead of v6.11-380 or similar. Depending on fft length on Radeon VII I've seen from <1% to 45.% difference; ~17.% average among all tested fft lengths.

[techn1ciaN]

Originally Posted by kriesel

What OS are you running?

If Windows, you may be paying a considerable penalty by running gpuowl v7.2 instead of v6.11-380 or similar. Depending on fft length on Radeon VII I've seen from <1% to 45.% difference; ~17.% average among all tested fft lengths.

Windows 11. A perfunctory check indicates 7747 μs/iter with 6.11 vs. the aforementioned 8031 μs/iter with 7.2, or about a 3.5% gain for 6.11.

I was reluctant to load 6.11 because I used tdulcet's PrimeNet script to register my 5700 XT as a PrimeNet "computer" and have been working towards qualifying it for cat 0 FTCs; I run 7.2 for the fused P-1 since most PRP FTCs seem to come without P-1 done. It occurs to me now, though, that I could simply haul the 6.11 results into the folder with my 7.2 instance and the script, let them be uploaded, and achieve the same result. Trying to hack the wrong solution into working is the mother of the right solution, or something...

That all said — I picked up a 100M-digit DC because I wanted to run some GPU DC, but couldn't get PrimeNet to register the low-cat "1 LL + 1 suspect" exponents that I bootlegged into PRP lines (this was on the 6th, before ZacHFX picked them up), and didn't want to load an LL-capable GPUOwl for the aforementioned reason. Now I see that this putative problem does not actually exist, so I will release

PRP=1,2,332242501,-1

and change it for

DoubleCheck=60550541,75,1
DoubleCheck=61304351,76,1
DoubleCheck=61326149,76,1
DoubleCheck=61373113,75,1
DoubleCheck=61422019,75,1

I apologize to Uncwilly for the trouble.