| T.R | Title | User | Personal
Name | Date | Lines |
|---|
| 15.1 | | BAYERN::WOLFF | Conformism is for little minds. | Wed Jun 26 1991 14:56 | 12 |
| We don't really have MFLOPS, all those figures are pretty meaningless anyhow
without specifying the floating point format etc.
1. The KAV30 does NOT support IEEE floating point operations
- therefore 0 MFLOPS using IEEE 32 or 64 bit operands
2. The KAV30 performs at 260 KWhetstones/Sec VAX G-Float
and 1700 KWhetstones/Sec VAX D-Float.
Julian.
|
| 15.2 | CSPI may have what you need ! | MGOI01::CSCHMIDT | Scio, Me Nil Scire | Wed Jul 17 1991 14:49 | 14 |
| If you're looking for an array processor on the VME, you may want to
contact CSPI in the US. They have a Q-Bus Array-Processor (of course,
we're not talking Q-Bus here) which is very reliable and relatively
cheap.
I know from one of my customers that they were supposed to develop one
for the VME-Bus, don't know about the status, though.
The address from the REALTIME-SOC spells:
CSP,Inc.
40 Linnell Circle
Billerica, MA 01812
Tel. (508) 662-0150
Cheers,
Christoph
|
| 15.3 | More VME supplier in TRIADD | UFHIS::WEISENBARTH | | Wed Jul 24 1991 10:19 | 16 |
| In the TRI/ADD program we work with several suppliers for VME array
processors together. However, as of today they are all focused to write
drivers for the TURBOchannel/VMEbus link under ULTRIX.
Since they are small companies and flexible, they may write ELN drivers
as well if you can provide funding for it. The list I have includes
CSPI (what was discussed before)
Analogics/CDA
SKY
For more information have a look in the TRIADD notes file
DECWSE::TRIADD.
Regards,
Wolfgang
|
| 15.4 | Same LINPACK MFLOPS as mVAX3400? | CESARE::OLOBARDI | Marco - Tech.OEM Mktg - Italy | Wed Dec 11 1991 19:02 | 23 |
| Hello,
again on MFLOPS.
I don't know if this is included in the extended report
mentioned in note 7.0; I have not copied the document.
We've been asked for a characterization of the KAV30 in terms
of floating point performance, single and double precision.
Since the KAV30 is based on rtVAX300 and this has the same chip of
mVAX3300 on board, I suppose this performance number is
the same, even if I don't know as regards cache, FPU, memory.
The figures that Digital gives for mVAX3400 are (source: microVAX
Perfo Summ. Sept 89 - 3100 et al.)
LINPACK
0.53 Single-precision (MFLOPS)
0.33 Double-precision (MFLOPS)
I don't know which format they used, maybe G and D?
Are these values the floating point measure of KAV30 as well or not?
Thanks for any help.
Ciao
Marco
|
| 15.5 | There is no IEEE floating point on KAV30 | BAYERN::WOLFF | Conformism is for little minds. | Fri Dec 13 1991 10:16 | 21 |
| RE: .-1
I say it again, I said it in .1 The problem is that most everything
if not actually everything on the VMEbus is IEEE floating point. Any NC machine
any other machine that produces floating point is IEEE. THE VAX DOES NOT DO
IEEE floating point. You can convert IEEE floating point to G-FLOAT VAX
datatype however, BUT THE CVAX DOES NOT IMPLEMENT G-FLOAT IN HARDWARE.
The IEEE floating point performance is nothing we should/can talk about. It will
be slow.
If your input source produces D-FLOAT you can use our FPU, and the
performance is reasonable good. But I won't say how much since the number is
meaningless in the VMEbus world. Additionally it is very hard to convert from
IEEE 64bit to D-FLOAT 64 bit so that you will loose any performace in the
conversion.
I hope this helps you understanding the problem and why asking for MFLOPS is
really not a valid question for the KAV30.
Julian.
|