How to do P-1 on a PRP assignment that has no P-1 work done?

[Uncwilly]

Originally Posted by drkirkby

Yes, but Canada, USA and South America on similar(ish) timezones.

You missed the point, we have folks outside that area. They can answer your question if they choose. But, I suspect they have chosen not to, because of your track record. BTW, USA and Canada together have time zones currently spanning 7 1/2 hours.

The position in the queue only got swapped for another category 0 exponent assigned the same day.

Why not just leave the assignments be.

BTW, I completed a triple check of one of the LL results you listed in another thread - M57906071. No less than 7 people got assigned LL tests, with 3 results, but at least two of them now agree.

Did you post in that thread that you were taking the exponent? Or is your queue so short that you were assigned it and it was completed in less than 24 hrs? It wasn't assigned when I updated the list with new exponents and you were not assigned it then. The result came in less than 24 hrs later.

[drkirkby]

Originally Posted by Uncwilly

You missed the point, we have folks outside that area. They can answer your question if they choose. But, I suspect they have chosen not to, because of your track record. BTW, USA and Canada together have time zones currently spanning 7 1/2 hours.
Why not just leave the assignments be.
Did you post in that thread that you were taking the exponent? Or is your queue so short that you were assigned it and it was completed in less than 24 hrs? It wasn't assigned when I updated the list with new exponents and you were not assigned it then. The result came in less than 24 hrs later.

1) Both axn and kriesel answered my question. kriesel was kind enough to do a better P-1 test for me.

2) Delaying the start of an exponent seemed quite reasonable, as I had some concerns about how the P-1 was performed. Had kriesel found a factor, I would have unreserved the exponent.

3) I answered the M57906071 query in the relevant thread.

[drkirkby]

Originally Posted by kriesel

For kicks, I put PFactor=0,1,2,103770473,-1,76,2 on a gpuowl run on Radeon VII GPU.

Do these Radeon VIIs have a reliability issue? A Google does not indicate a particular problem, but there are an awful lot on eBay for "parts only". Faulty ones are selling for 50% more than I paid for 26-core Xeons. No video output seems a common feature. If the faulty ones were a sensible price, I would be willing to take a chance and hope they worked for computational tasks, and not worry about the lack of video. But the prices are crazy for non-working items. $199 seemed cheap, until I realised that was only for the box!

These are out of my league.

[kriesel]

The time to buy Radeon VIIs was about a year ago; new with 2 year warranty for $600-750 each. Later I picked some up used for about the same price range. Each Radeon VII, properly tuned with a good gpuowl version, can match PRP throughput of your $5000 dual-Xeon system; a 104M PRP / day. The case or mining style frame, a large PSU, a modest cpu and motherboard, a couple DIMMS, add maybe ~$100/GPU. Or more recently, a diskless Xeon Phi 7250 based system cost $700 used (years ago was ~$5k new). Some of us got lucky and the vendor shipped 7250s when we ordered $500 used 7210 systems. A pair of those provide PRP throughput about equal to your $5k system.

The current pricing for used working Radeon VIIs is remarkable. I've seen sales for up to $2400, 3-4 times new cost. This week was ~$1600. Broken no-display, no-OpenCL-compute, or even not-seen-by-the-OS (completely bricked) sell for almost original new price. Perhaps there are shops that can repair some fraction, given enough carcasses. Coincidentally, the manufacturer I have a pending warranty claim with is not repairing or replacing in-warranty Radeon VIIs, but offering either an RX5700XT (~45% of PRP or P-1 performance) or original-retail-price refund, which might buy 1/4-1/3 of an equal-performance GPU if we could find one in stock. (~RX6900XT)

[LaurV]

They are used for mining (arguably, still the best card). People buy them, OC them, fry them, roast them, then want to sell them for more money than they bought them new. Don't buy, unless you know exactly what you are doing, and are willing to dirty your hands with soldering (and have the tools - a simple soldering iron won't do it, the substrate of that PCB is freaking dissipative, you will not be able, or will need ages, to heat it till the tin melts). Talking from own experience (working in an electronics factory).
(edit: we crossposted, I was replying to doctor).

[VBCurtis]

Originally Posted by drkirkby

Faulty ones are selling for 50% more than I paid for 26-core Xeons.

I think I'll start a count of posts where the famous 26-core Xeons are mentioned.

One.

[Uncwilly]

[kriesel]

26... Is that considered a lot? I remember when I thought 12 was. Or dual-quad-core Xeon was a lot of cpu power. Or dual-pentium-mmx-200. Or 386-33 with two levels of caching (subsequently upgraded to a TI 486/66 core & Cyrix Fasmath coprocessor). Or 8 MHz NEC 70108 in an 8088's socket with the PCB of an original IBM PC solder-iron-hacked to split the clock signals out and put faster 74LS TTL chips in certain locations and overclock the base design ~50-60% while preserving the 14.31818 MHz /3 and /4 frequencies needed for IO bus and timing.