Reply by glen herrmannsfeldt●November 3, 20132013-11-03
Tim Wescott <tim@seemywebsite.really> wrote:
>> Tim Wescott <tim@seemywebsite.please> wrote:
(snip, I wrote)
>>>> It is not so hard to design an add-on coprocessor, but getting useful
>>>> work from one is somewhat harder. First you have to remember Amdahl's
>>>> law. Second, you have to make sure that you are not I/O limited.
>>> It sounds like cold fusion to me.
>> It isn't quite that bad, but it isn't so easy, either.
> << comments snipped >>
> I'm not sure what you're reading into what the OP said, but he's
> promising more than 1e3 improvement in speed just by hooking some
> gizmo up to a USB port on "any PC in the world", without specifying
> what its going to do or how.
Rereading the original post, it doesn't mention at all what problem
it applies to. More specifically, it doesn't say that it will speed
up all problems.
How fast can most machines do 8 bit adds these days? It isn't
hard to build an array that can do more than 1000 of them in
the time another system can do one. If your problems need only
8 bit adds, it might work.
You also have to be lucky that the needed I/O rate is low enough.
Though USB 3.0 isn't so bad.
> What I read into that is "here be magic, just give me money".
Need some luck, too.
-- glen
Reply by Tim Wescott●November 2, 20132013-11-02
On Sat, 02 Nov 2013 20:33:42 +0000, glen herrmannsfeldt wrote:
> Tim Wescott <tim@seemywebsite.please> wrote:
>
> (snip, I wrote)
>
>>> Over the years there have been a number of tries at building and
>>> selling co-processors for hardware acceleration. Many companies
>>> designing and selling such are now gone.
>
>>> Some that are still around are drccomputer and timelogic.
>
>>> It is not so hard to design an add-on coprocessor, but getting useful
>>> work from one is somewhat harder. First you have to remember Amdahl's
>>> law. Second, you have to make sure that you are not I/O limited.
>
>> It sounds like cold fusion to me.
>
> It isn't quite that bad, but it isn't so easy, either.
<< comments snipped >>
I'm not sure what you're reading into what the OP said, but he's
promising more than 1e3 improvement in speed just by hooking some gizmo
up to a USB port on "any PC in the world", without specifying what its
going to do or how.
What I read into that is "here be magic, just give me money".
--
Tim Wescott
Wescott Design Services
http://www.wescottdesign.com
Reply by glen herrmannsfeldt●November 2, 20132013-11-02
Tim Wescott <tim@seemywebsite.please> wrote:
(snip, I wrote)
>> Over the years there have been a number of tries at building and selling
>> co-processors for hardware acceleration. Many companies designing and
>> selling such are now gone.
>> Some that are still around are drccomputer and timelogic.
>> It is not so hard to design an add-on coprocessor, but getting useful
>> work from one is somewhat harder. First you have to remember Amdahl's
>> law. Second, you have to make sure that you are not I/O limited.
> It sounds like cold fusion to me.
It isn't quite that bad, but it isn't so easy, either.
First you have to have a problem that is mostly fixed point add
and subtract, maybe some multiplies, too. A favorite one is
dynamic programming pattern matching algorithms, such as are
used for comparing DNA and protein sequences. That is what
timelogic sells, and seems to be still in business.
FIR filters might not be so bad, either. Using an FPGA with lots
of multiplier blocks would help.
If your problems do have a lot of communication between the host
and the coprocessor, you need a fast link. DRCcomputer has some
that fit in the processor socket of an existing multiprocessor
board, such as Opteron. Others are PCIe, which also has a pretty
good data transfer rate. That allows for problems with more I/O,
but they still have to fit. FPGAs are getting pretty big, though,
and for reasonably prices.
-- glen
Reply by Tim Wescott●November 2, 20132013-11-02
On Sat, 02 Nov 2013 09:15:27 +0000, glen herrmannsfeldt wrote:
> Dan Wawa <danwawa6@gmail.com> wrote:
>> Hi, I have designed a microprocessor that works as a USB device that I
>> call DorQ. Its function is to speed up the processing speed of any
>> computer that it is connected to by a factor of several thousands of
>> times.
>
> Over the years there have been a number of tries at building and selling
> co-processors for hardware acceleration. Many companies designing and
> selling such are now gone.
>
> Some that are still around are drccomputer and timelogic.
>
> It is not so hard to design an add-on coprocessor, but getting useful
> work from one is somewhat harder. First you have to remember Amdahl's
> law. Second, you have to make sure that you are not I/O limited.
>
> \-- glen
It sounds like cold fusion to me.
--
Tim Wescott
Control system and signal processing consulting
www.wescottdesign.com
Reply by glen herrmannsfeldt●November 2, 20132013-11-02
Dan Wawa <danwawa6@gmail.com> wrote:
> Hi, I have designed a microprocessor that works as a USB device
> that I call DorQ. Its function is to speed up the processing
> speed of any computer that it is connected to by a factor of
> several thousands of times.
Over the years there have been a number of tries at building
and selling co-processors for hardware acceleration. Many companies
designing and selling such are now gone.
Some that are still around are drccomputer and timelogic.
It is not so hard to design an add-on coprocessor, but getting
useful work from one is somewhat harder. First you have to
remember Amdahl's law. Second, you have to make sure that you
are not I/O limited.
\-- glen
Reply by Robert Miles●November 2, 20132013-11-02
On Wednesday, October 16, 2013 12:17:19 PM UTC-5, Dan Wawa wrote:
> Hi, I have designed a microprocessor that works as a USB device that I ca=
ll DorQ.
I don't know any such company, but a few comments anyway:
It looks best to patent this design before offering it to anyone.
Your description leaves out a few important details, such as what types of =
programs it will speed up that much. Many high-end graphics boards can alr=
eady speed up programs by a factor of a few thousand, but only for programs=
that have large numbers of sections that can be run in any order or even a=
ll at once because they do not affect each other (parallel processing, ofte=
n using OpenCL). If this fits your device, offering an OpenCL compiler for=
it would help it sell.
Also, will it contain a large memory, or must it reach such a memory throug=
h the USB connection (which is rather slow for this purpose)?
If it can speed up all programs by a factor of several thousand, even those=
where all sections must be run one after the other, I'd expect Intel, AMD,=
and any other microprocessor makers you can find to be very interested - b=
ut also with large budgets to fight legal battles for or against you. The =
companies making quantum computers are also likely to be interested, since =
they are already working on especially fast computers.
Reply by Dan Wawa●October 16, 20132013-10-16
Hi, I have designed a microprocessor that works as a USB device that I call=
DorQ. Its function is to speed up the processing speed of any computer tha=
t it is connected to by a factor of several thousands of times. I call it t=
he concept of Parallel Bus Processing. My design is still in VHDL source co=
de and I need a partner to help fabricate/manufacture and market this new k=
ind of USB device. I repeat, this is a completely new USB device that will =
work with any PC in the world. Would you be interested in partnering with m=
e in this venture? Please contact me with your yes or no response or any id=
eas about companies that may accept designs from partners as part of their =
product lines. You may contact me at loumbut5@hotmail.com