Re: best chipset/cpu

[Riccardo Sibilia <riccardo.sibilia@gmx.ch>]

Ok, I'm not on the bussines of answering my own questions, but I though it 
could be of help for some other person.... the threat "NMBCLUSTERS, 
NKMEMPAGES, kernel panics, too much memory?" actually answers my questions. 
The dmesg that Paul reported states:

pchb0 at pci0 dev 0 function 0 vendor "Intel", unknown product 0x2540 rev 0x02
vendor "Intel", unknown product 0x2541 (class undefined, unknown subclass 
0x00, rev \ 0x02) at pci0 dev 0 function 1 not configured
ppb0 at pci0 dev 2 function 0 vendor "Intel", unknown product 0x2543 rev 0x02
pci1 at ppb0 bus 1

and the 0x2540 etc. are Intel E7500 chipset components.

It's not yet known to the OS but it work.

The RNG support is for sure not yet there for know... but that's no reason not 
to start work on that beast. I'll take a look at RNG when the machines are 
delivered. Can't be too diverent from the 860, I think.

Cheers
Rick


On Wednesday 31 July 2002 07:19, Riccardo Sibilia wrote:
> On Tuesday 30 July 2002 23:57, Theo de Raadt wrote:
> > > http://openbsd.org/crypto.html#hardware
> > >
> > > If you care about crypto throughput above all else, a general-purpose
> > > CPU will always be slower than a dedicated hardware solution.
> >
> > Not neccessarily.
> >
> > Small operations may be impacted by the higher latency of queuing them
> > off to another device.
>
> Yes... and since we are implementing "our own algorithm" which is quite
> optimized on the x86 CPUs, we are quite confident we can get a better
> overall throughput than any 3DES dedicated board. If not on the gateway
> (OpenBSD-based) then for sure on the VPN clients (M$ based).
>
> Just a test question: which algorithm is pretty well tested, has lot of
> cryptoanalytical results (and well known and documented set of weak keys),
> requires a small footprint, no tables or s-boxes and gets fast as hell if
> you can perform 16x16 bit multiply quickly? ... no, it's not twofish :-)))
>
> Still I/O and memory access times and latency is an issue and I still do
> not know which one of the newest chipsets is not going to be a pain in the
> ass to use....
>
> Intel i850? (i845, etc.)
> Intel E7500?
> ServerWorks GC?
> AMD 756? (well, I know about this one....)
>
> thx
>
> ciao
> Rick