WOW!

[ET_]

A nice tool to enhance linear algebra step...

RevoDrive

Luigi

[firejuggler]

couple it with http://www.tomshardware.com/news/int...sor,14002.html
and it goes even faster

[Batalov]

Originally Posted by ET_

A nice tool to enhance linear algebra step...

RevoDrive

Luigi

How is going to enhance linear algebra step?
Reading the relations and the matrix from the disk takes negligible time.

[ET_]

Originally Posted by Batalov

How is going to enhance linear algebra step?
Reading the relations and the matrix from the disk takes negligible time.

Does it? Wasn't there a step highly disk-i/o involved? So heavy that CPU-time couldn't reach constant 100%?

If I contradicted myself, then I contradicted myself...

Luigi

[Batalov]

You meant filtering then. It used to take days a few years ago, but after jasonp's optimizations - it, too, takes negligible time now.

Maybe - for some unique, monstrousfiltering job (e.g. M1061) it will take a day or two again, but the rest of the stages will take CPU-years, so still negligible proportionally.

[xilman]

Originally Posted by Batalov

You meant filtering then. It used to take days a few years ago, but after jasonp's optimizations - it, too, takes negligible time now.

Maybe - for some unique, monstrousfiltering job (e.g. M1061) it will take a day or two again, but the rest of the stages will take CPU-years, so still negligible proportionally.

IMO, filtering is (a technique for) linear algebra.

My understanding of "LA" is that it is the process of finding linear dependencies over F₂ within the collection of relations found by sieving. The entire process could be performed by Gaussian elimination, blocked Lanczos or blocked Wiedemann (and doubtless other algorithms) but efficiency concerns indicate that filtering is a useful first stage.

Paul

[Batalov]

We are in violent agreement!?

the toe bone connected
to the foot bone,
and the foot bone connected
to the ankle bone,
and the ankle bone connected
to the leg bone.
Oh mercy how they scare!

The point is that i/o is negligible in all stages. If anywhere, it is noticeable in filtering.

Incidentally, in multi-core BL cpu load is not reaching 100% for an entirely different reason. (Throwing gas on fire...)

[jasonp]

Back in 2000, I wondered how much work it would have been to put gobs of DRAM onto a PCI board, to function as (volatile) backing store for the DRAM on the motherboard. Now the kind of memory capacity I considered unthinkable in 2000 costs $10.

The page says their product is targeted at the gaming market, but I can't see how they'd get the price of such a board to be less than that of the entire machine it's plugged into.

[LaurV]

Originally Posted by jasonp

Back in 2000, I wondered how much work it would have been to put gobs of DRAM onto a PCI board, to function as (volatile) backing store for the DRAM on the motherboard. Now the kind of memory capacity I considered unthinkable in 2000 costs $10.

The page says their product is targeted at the gaming market, but I can't see how they'd get the price of such a board to be less than that of the entire machine it's plugged into.

Technically they all use cheap serial nand flash (as contrary to the much faster and expensive parallel nor flash), which is quite cheap now. A microSD with 32GB is below 25 bucks here. I imagine having 10 or 20 like that in a small box, in parallel, is no big deal. We had a customer who wanted us to produce ssd's but my boss did not take the order, I don't know why.

[pinhodecarlos]

BTW, Happy Birthday Luigi!!!

[ET_]

Originally Posted by pinhodecarlos

BTW, Happy Birthday Luigi!!!

Thank you!