Potentially suspect results (mine)
 

Can you mark these PRP as potentially suspect please. They are from a recent GPU undervolt and it just failed a DC. I've dialed back the undervolt and am now testing with different DCs.



[URL]https://www.mersenne.org/report_exponent/?exp_lo=77988359&full=1[/URL]
[URL]https://www.mersenne.org/report_exponent/?exp_lo=84080669&full=1[/URL]


edit: For future reference these were done with gpuowl 4.7-5b01b65 at 4608K.

Which DC failed?
(note that the residue is normally different between "type-4" and "type-1")

[QUOTE=M344587487;499532]Can you mark these PRP as potentially suspect please. They are from a recent GPU undervolt and it just failed a DC. I've dialed back the undervolt and am now testing with different DCs.



[URL]https://www.mersenne.org/report_exponent/?exp_lo=77988359&full=1[/URL]
[URL]https://www.mersenne.org/report_exponent/?exp_lo=84080669&full=1[/URL]


edit: For future reference these were done with gpuowl 4.7-5b01b65 at 4608K.[/QUOTE]

[QUOTE=M344587487;499532]Can you mark these PRP as potentially suspect please. They are from a recent GPU undervolt and it just failed a DC. I've dialed back the undervolt and am now testing with different DCs.



[URL]https://www.mersenne.org/report_exponent/?exp_lo=77988359&full=1[/URL]
[URL]https://www.mersenne.org/report_exponent/?exp_lo=84080669&full=1[/URL]


edit: For future reference these were done with gpuowl 4.7-5b01b65 at 4608K.[/QUOTE]

Queued the one that wasn't assigned.

Apologies I didn't notice replies until now. You're right that I was looking at the residues and saying they were suspect because they didn't match. If the residue types are not comparable how can we verify that the result was correct? Is it fair to say that we've done two first time PRP tests and that someone still needs to come in with a DC to match one of them?

[QUOTE=M344587487;499721]Apologies I didn't notice replies until now. You're right that I was looking at the residues and saying they were suspect because they didn't match. If the residue types are not comparable how can we verify that the result was correct? Is it fair to say that we've done two first time PRP tests and that someone still needs to come in with a DC to match one of them?[/QUOTE]

A residue of some type can be verified only by a double-check of the same type. In practice this works like this: mprime / prime95 can verify any type. Once a first PRP is in, the server will emit a DC assignment of matching type for verification.

The "trouble maker" here is gpuowl, who now generates "type-4" instead of the more common "type-1" PRP. (but there is some reason for why gpuowl does that).

Thanks for the explanation, in that case I've dequeued the remaining DC work and my results are no more suspect than any other that's never been checked. I'll look for a type 4 DC.

[QUOTE=preda;499726]The "trouble maker" here is gpuowl, who now generates "type-4" instead of the more common "type-1" PRP. (but there is some reason for why gpuowl does that).[/QUOTE]

For some reason, Primenet marks your first-time type-1 result for [M]M77988359[/M] as "Reliable", but not the type-4 from a later version of the same software.

I queued up the double-check, but it won't begin for a couple of days. I will set it to target the type-4 result, which is from newer code and presumably in greater need of verification.

[QUOTE=GP2;499748]For some reason, Primenet marks your first-time type-1 result for [M]M77988359[/M] as "Reliable", but not the type-4 from a later version of the same software.

I queued up the double-check, but it won't begin for a couple of days. I will set it to target the type-4 result, which is from newer code and presumably in greater need of verification.[/QUOTE]

The type-1 result there is from mprime, not from GpuOwl (because it has offset!=0, while GpuOwl has offset==0). So it's the mprime result that's marked "reliable".

Thanks for the double-check of the type-4 result from GpuOwl, that's useful.

[QUOTE=GP2;499748]For some reason, Primenet marks your first-time type-1 result for [M]M77988359[/M] as "Reliable", but not the type-4 from a later version of the same software.

I queued up the double-check, but it won't begin for a couple of days. I will set it to target the type-4 result, which is from newer code and presumably in greater need of verification.[/QUOTE]
Thanks, I can't easily find a wavefront type 4 to DC so a verify whenever convenient could save a lot of cycles if my HW is spitting out junk.

[QUOTE=GP2;499748]For some reason, Primenet marks your first-time type-1 result for [M]M77988359[/M] as "Reliable", but not the type-4 from a later version of the same software.

I queued up the double-check, but it won't begin for a couple of days. I will set it to target the type-4 result, which is from newer code and presumably in greater need of verification.[/QUOTE]

Do you do this via manual assignments? I wanted to double check [M]M50012323[/M] again as it was a mismatch, but that computer has since done two more successful DCs, so I wanted to throw that another computer. Or should I not worry about it?

[QUOTE=irowiki;499767]Do you do this via manual assignments? I wanted to double check [M]M50012323[/M] again as it was a mismatch, but that computer has since done two more successful DCs, so I wanted to throw that another computer. Or should I not worry about it?[/QUOTE]

It isn't considered a good idea to double check your own results, since an independent double-check will probably get done regardless.

I wouldn't worry about it. The other computer on the first-time check has a track record of bad results, around 20% bad.