gpuOwL: an OpenCL program for Mersenne primality testing

[preda]

Originally Posted by kriesel

If 3570 meant this, perhaps the igp suddenly began to show up as an available OpenCL device. (HD2500 listed as a device in gpuowl -h output?)

Nice detective work! (yes, that sounds like a reasonable explanation)

[ixfd64]

I noticed something a little weird. A factor for 104011507 was recently found by P-1 with B1=5,000,000. However, mersenne.ca shows that the largest factor of k is 5,902,627 — which is above the B1 value. Is gpuOwl somehow able to find factors outside the given bounds?

Or at least I'm assuming this factor was found with gpuOwl given how long it would take on a CPU. I know that the Brent–Suyama extension can be used to find factors outside given ranges, but it's my understanding that it only applies when stage 2 is used.

[preda]

Originally Posted by ixfd64

I noticed something a little weird. A factor for 104011507 was recently found by P-1 with B1=5,000,000. However, mersenne.ca shows that the largest factor of k is 5,902,627 — which is above the B1 value. Is gpuOwl somehow able to find factors outside the given bounds?

Or at least I'm assuming this factor was found with gpuOwl given how long it would take on a CPU. I know that the Brent–Suyama extension can be used to find factors outside given ranges, but it's my understanding that it only applies when stage 2 is used.

It is simply misreporting from GpuOwl: when it finds a P-1 factor, the part that writes the result doesn't know anymore whether it was found in stage1 or in stage2. It could write the result with both B1 and B2, but this is interpreted by the server as implying "found in stage2", and results in allocating a larger credit. So to be conservatory relative to the credit, only B1 is reported (which implies "found in stage1" and results in a bit lower credit).

I should probably fix this when I get to it.

You are correct, there is no way to find in stage1 a factor larger than B1 (there is no BS analogue for stage1); in addition, GpuOwl does not use BS in stage2.

[ixfd64]

I see, thanks for the explanation. On a related note, are there any plans to implement the Brent–Suyama extension in the future?

[preda]

Originally Posted by ixfd64

I see, thanks for the explanation. On a related note, are there any plans to implement the Brent–Suyama extension in the future?

Not ATM. Based on the information available, it is more efficient to run with a higher B2 than to use BS (for the same amount of compute in P2).

[Xyzzy]

This is with the default fan and power profile.

2020-11-13 22:00:58 GeForce RTX 3080-0 OpenCL compilation in 1.32 s
2020-11-13 22:00:59 GeForce RTX 3080-0 77936867 OK        0 loaded: blockSize 400, 0000000000000003
2020-11-13 22:01:01 GeForce RTX 3080-0 77936867 OK      800   0.00%; 1925 us/it; ETA 1d 17:41; 1579c241dc63eca6 (check 0.80s)
2020-11-13 22:07:31 GeForce RTX 3080-0 77936867 OK   200000   0.26%; 1954 us/it; ETA 1d 18:12; f0b04b45b0855bd2 (check 0.81s)
2020-11-13 22:14:05 GeForce RTX 3080-0 77936867 OK   400000   0.51%; 1966 us/it; ETA 1d 18:21; c03f94396a5aa29e (check 0.81s)
2020-11-13 22:20:37 GeForce RTX 3080-0 77936867 OK   600000   0.77%; 1954 us/it; ETA 1d 17:59; b9decd65ca71b629 (check 0.81s)

[moebius]

Originally Posted by Xyzzy

GeForce RTX 3080

Unfortunately still not better than the RTX 2080 TI.But, the RTX 3080 TI should appear on the market in January 2021
Ranking List

[moebius]

Originally Posted by moebius

The launch is likely to be on November 16, 2020 but with up to 12.5 TFLOPs FP64?

real specifications http://www.digitaljournal.com/pr/4879636

price: around US-$ 6400

[M344587487]

https://www.servethehome.com/amd-rad...-gpu-launched/


Based on the slides it should be 56-65% the cost of an A100, your number looks about right. The chance of a consumer version is ~0%, but there may be a pro variant. The poorly binned dies have to go somewhere, maybe there'll be an MI90 instead which would be a sad development.

[Xyzzy]

RX 5500XT

2020-11-20 13:08:06 gfx1012-0 OpenCL compilation in 2.57 s
2020-11-20 13:08:07 gfx1012-0 79334869 OK        0 loaded: blockSize 400, 0000000000000003
2020-11-20 13:08:11 gfx1012-0 79334869 OK      800   0.00%; 3287 us/it; ETA 3d 00:26; 0b08957c2d2b3130 (check 1.37s)
2020-11-20 13:19:01 gfx1012-0 79334869 OK   200000   0.25%; 3279 us/it; ETA 3d 00:05; 79983c264372e180 (check 1.35s)
2020-11-20 13:29:56 gfx1012-0 79334869 OK   400000   0.50%; 3272 us/it; ETA 2d 23:45; 76c411b78b92c115 (check 1.35s)
2020-11-20 13:40:52 gfx1012-0 79334869 OK   600000   0.76%; 3272 us/it; ETA 2d 23:34; 2e03b6c89817762d (check 1.35s)

[moebius]

Originally Posted by Xyzzy

RX 5500XT

https://mersenneforum.org/showpost.p...postcount=2547
Thank you, a forum user has slightly better values ​​for the AMD RX 5500 XT, but I think Sapphire is the measure of all things. The exponent is also not exactly the same.