Did Amazon just join GIMPS?

[ixfd64]

If you look at the Top Producers page, you'll see that a user named "Amazon EC2" as produced over 40,000 GHz-days of results in less than 90 days. This suggests that Amazon has joined GIMPS with their cloud computing farm! Either that, or we have a GIMPSter with a lot of money. If it's the former case - and I really think it is - then it's pretty nice to have a big player crunching for us!

[science_man_88]

Originally Posted by ixfd64

If you look at the Top Producers page, you'll see that a user named "Amazon EC2" as produced over 40,000 GHz-days of results in less than 90 days. This suggests that Amazon has joined GIMPS with their cloud computing farm! Either that, or we have a GIMPSter with a lot of money. If it's the former case - and I really think it is - then it's pretty nice to have a big player crunching for us!

if I did the correct math that means an average of 444-445 GHz/day, an enjoyable pace.

[Batalov]

Oh, so the contributions from Los Alcoholicos or Jack The Ripper does not deserve the reader's attention, ...just the Amazon? We saw it on the internet - so it must be true.

[LiquidNitrogen]

Originally Posted by science_man_88

if I did the correct math that means an average of 444-445 GHz/day, an enjoyable pace.

Buckeye has been averaging 25+ Ghz-days per day on one of our 5.0 GHz i7-2600K machines with 4 cores and no hyperthreading.

That would be 17.76 systems to generate 444 GHz-days/day.

My guess would be it's no more than 40 Amazon machines sitting idle.

[ixfd64]

One thing to note is that Amazon's farm uses Xeon chips, which have a lower clock rate. As such, it may take up to two months for a single core to finish a ~50M exponent. This would explain why most of this user's results are double checks.

I guess we'll get a better image in a couple of months.

Edit: the Top Producers page shows that "Amazon EC2" started returning results WITHIN the last 90 days. It does not mean that the user returned 44,000 GHz-days in 90 days. Once the asterisks disappear, we'll know that 90 days has passed since the first result.

[Xyzzy]

Two GPU setups net you 540GHz/days per day.

[TheJudger]

Originally Posted by Xyzzy

Two GPU setups net you 540GHz/days per day.

But only for Trial Factoring.
Primenet needs more horsepower on Lucas Lehmer and Double Checks.
Xyzzy: how much throughput has your setup (2x GTX 570?) for LL/DCs?

Oliver

[Christenson]

Mr P-1 and I will also tell you that the most candidates will be eliminated right now for a CPU with P-1 factoring; it doesn't obviate the LL tests, but, like TF, it removes some of them, and I am averaging noticeably fewer GHz-Days per factor found (and therefore LL eliminated) than I would if I did both LL and LL-D on the same exponents.

[davieddy]

Originally Posted by Christenson

Mr P-1 and I will also tell you that the most candidates will be eliminated right now for a CPU with P-1 factoring; it doesn't obviate the LL tests, but, like TF, it removes some of them, and I am averaging noticeably fewer GHz-Days per factor found (and therefore LL eliminated) than I would if I did both LL and LL-D on the same exponents.

Yes, which is precisely why P-1 is worth doing (memory permitting)
before embarking on an LL test.

Originally Posted by TheJudger

But only for Trial Factoring.
Primenet needs more horsepower on Lucas Lehmer and Double Checks.
Xyzzy: how much throughput has your setup (2x GTX 570?) for LL/DCs?

Oliver

I posted a "Puzzle" thread entitled "TF factoring limits" specifically
inviting a good answer to "How far can GPUS TF usefully?".
CRGreathouse suggested that the answer depended partly on how
much GPU firepower was out there.
Since the firepower is now considerable, I think it is best employed
TFing a few more bits , keeping just ahead of the LL wavefront
(currently crawling through the 53M range).

I would argue that all factoring done so far above 60M is
effectively worthless as far as finding another MP is concerned,
in the sense that it is a small fraction of the work needed before
an LL test becomes clearly worthwhile.

Fully aware that one is "doing the lottery", it is incentivizing when
embarking on an LL test to know that
a) It couldn't be cheaply avoided
b) The chance of being prime has been enhanced by 1.5*(extra bits)/70%.

David

[Xyzzy]

Xyzzy: how much throughput has your setup (2x GTX 570?) for LL/DCs?

So little that we would not waste the electricity.

[NBtarheel_33]

Originally Posted by Xyzzy

So little that we would not waste the electricity.

But I thought that folks were having success with LL testing on these GPUs with CUDALucas. I can remember some extremely quick LL tests being completed in the 80M range, and also one at 123M. What happened?