[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <CAOLP8p5ubRtJ6dfZKetK4x+SfHkxMD71GG9kk3g5NCEdbEH7rA@mail.gmail.com>
Date: Sun, 19 Jul 2015 07:27:15 -0700
From: Bill Cox <waywardgeek@...il.com>
To: "discussions@...sword-hashing.net" <discussions@...sword-hashing.net>
Subject: Re: [PHC] Bandwidth hardened algorithms?
On Thu, Jul 16, 2015 at 4:38 PM, Dmitry Khovratovich <khovratovich@...il.com
> wrote:
> I see your point. But don't you overestimate the ASIC cost? As there is
> little logic beyond the memory itself, the parallel chip can be as cheap as
> the 32 GB unit, i.e. 180$
>
> Dmitry
>
I am underestimating the ASIC cost in the extreme case where a buyer buys
many millions of dollars worth. A lot of the pricing depends on volume.
They'd have to buy many millions of dollars worth to get the best pricing,
but I suppose BitCoin mining might support that.
High bandwidth ASICs and the systems they go into are tricky. For example,
Achronix only sells their FPGAs to a few big customers. You can't buy
one. When you have 10 GiB/s I/O pins, you can't just let anybody try to
build a board using them. Instead you have to get involved with the
customer in their board design. I heard that the Rocket I/O project at
Xilinx slowed down their FPGA development to a point that it did major
damage to the company, and these I/Os are 3X slower than the Achronix I/Os
per pin.
I think DRAM pricing shows a good example of how cheap high bandwidth
devices can be. The highest bandwidth devices are quite a bit more
expensive. It is not clear how much the actual RAM costs vs the I/Os in
these parts.
Bill
Content of type "text/html" skipped
Powered by blists - more mailing lists