3rd party DLT drive performance
Kevin Johnson
if it's actually 1.25MB/sec uncompressed? I realize all DLT drives are made by
the same manufacturer (was DEC, now Quantum) but I heard from a 3rd party that
a customer who tested the non-SGI drives found they were performing at a speed
closer to that of DAT. I'm using an Indy which SGI doesn't sell DLT's for.
Thanks in advance for your reply.
Kevin
--
Kevin Johnson
ke...@reachinfo.com
Flo Rueb
I just have a question regarding the line of MIPS processors. Which of
the one: R4400, R4600 or R4700 is faster. Is it just the clock speed.
So is the 150MHz R4400 faster than the 133MHz R4600. If so, what's the
difference between the processors. I believe that the fastest one out
of those is the 200MHz R4400. Why would the continue the line to 4600
and 4700 then?
Thank you.
Martin Knoblauch
writes:
At comparable clock speeds, the R4400 offers a lot more FP power
than the R4600 (the R4600 might have a slight speed advantage on
INTEGER) at a visibly higher price. I can't say much about the
R4700, but it is said to bridge the FP-gap between R4400 and R4600
(probably also the price gap).
The speed differences between R4400 and R4600/R4700 are mainly
due to differences in the cache architecture and the FPU pipeline.
For more detailed technical information you should look at:
you will find a lot of interesting MIPS information there.
As the R4600's are cheaper than the R4400, it makes sense to
use them for applcations that are not FP-hungry.
Hope this helps a bit
Martin
--
+---------------------------------+-----------------------------------+
|Martin Knoblauch | Silicon Graphics GmbH |
|Application Center | Am Hochacker 3 - Technopark |
|Silicon Graphics Computer Systems| D-85630 Grasbrunn-Neukeferloh, FRG|
| | Phone: (+int) 89 46108-179 or -0 |
| | Fax: (+int) 89 46108-190 (-222) |
+---------------------------------+-----------------------------------+
|Network: <kn...@munich.sgi.com> | V-Mail: 5-8935 | M/S: IMU-315 |
+---------------------------------------------------------------------+
Dave Olson
| Has anyone measured the performance of 3rd party DLT drives under 5.3 OS to see
| if it's actually 1.25MB/sec uncompressed? I realize all DLT drives are made by
| the same manufacturer (was DEC, now Quantum) but I heard from a 3rd party that
| a customer who tested the non-SGI drives found they were performing at a speed
| closer to that of DAT. I'm using an Indy which SGI doesn't sell DLT's for.
If it's performing that badly, they have something wrong with their
setup, or are doing something silly likely writing really small blocks
and not running in buffered mode. The single-ended version of the
DLT should work just fine on an Indy at 5.3.
--
The most beautiful things in the world | Dave Olson, Silicon Graphics
are those from which all excess weight | http://reality.sgi.com/employees/olson
has been removed. -Henry Ford | ol...@sgi.com
sum...@fnalnj.fnal.gov
fl...@ix.netcom.com (Flo Rueb) writes:
>Hi everyone!
>
>I just have a question regarding the line of MIPS processors. Which of
>the one: R4400, R4600 or R4700 is faster. Is it just the clock speed.
>So is the 150MHz R4400 faster than the 133MHz R4600. If so, what's the
>difference between the processors. I believe that the fastest one out
>of those is the 200MHz R4400. Why would the continue the line to 4600
>and 4700 then?
>
>Flo
>fl...@ix.netcom.com
**********************************************************************
COMPARISON OF R4400, R4600, and R4700 CPUs
R4600 CYCLES R4700 CYCLES R4400 CYCLES
SINGLE DOUBLE SINGLE DOUBLE SINGLE DOUBLE
FLOATING POINT ADD 4 4 4 4 4 4
FLOATING POINT MULTIPLY 8 8 5 6 7 8
FLOATING POINT DIVIDE 32 61 32 61 23 36
FLOATING POINT SQRT 31 60 31 60 54 112
OVERLAPPED
FLOATING POINT MULTIPLY 6 6 4 5
INTEGER MULTIPLY 10 12 8 10 10 20
INTEGER DIVIDE 42 74 42 74 69 133
The R4700 doe floating point multiplies in as few as 4 clock cycles, 33% faster
than the R4600 at 6 clock cycles, The R4700 is available in 4 speeds; 100,
133, and 150 MHz as is the R4600 plus a new R4700 speed of 175 MHz.
In all cases, SHIFT is much faster than DIVIDE when dividing by a power of 2.
SHIFTS may also be used to prevent INTEGER and FLOATING POINT MULTIPLIES from
interfering (both use the same floating point silicon in the R4600/R4700). For
example, any multplty which involves a number which is the sum or difference of
two powers of 2, only requires two SHIFTs and one ADD or SUBTRACT. This is
faster than a MULTIPLY even if there is no conflict. It may be particulary
helpful when calculating array indices of matrices. The numbers meeting this
requirement are
2 3 4 5 6 7 8 9 10 12 14 15 16 17 18 20 24 28 30 31 32 33 34 36 40 48 56 60 ...
The R4700 is pin and instruction compatible with the R4600 and R4400PC. Its "set
associative" 16kB caches let it run fairly fast without an external cache.
Both the R4600 and R4700 are less expensive than the R4400.
Now one might speculate that if the INDY CPU card had sockets, one might pull
the 133MHz R4600 out and just plug in the 150MHz R4700 along with a 50MHz clock
chip to replace the old 44 MHz clock chip. The Sclock divisor of 3 read from
the EEPROM chip remains unchanged. If you want the 175MHz R4700 leave the
44MHz clock chip alone and burn a new EEPROM chip with an Sclock divisor of 4
rather than 3. Turn mode bit 5 off and mode bit 6 on to increase the divisor.
Even greater gains might be possible with 100MHz R4000/R4600 CPU boards.
The INDY will allow an external clock of 50 MHz, but the R4600/R4700 permits
higher speeds. A CPU card secondary cache might be able to run faster than
50 MHz, as long as the access off the CPU card is held to 50 MHz.
Internal
Clock MHz 100 133 150 175
Sclock /2 50 66 75 87 External Clock MHz
Divisor /3 44 50 58
/4 44
The ratings of the INDY with the 133MHz R4600SC CPU recently increased.
Does anyone know what was improved? Is the change incorporated in new versions
of the FORTRAN and C compilers?
OLD NEW CHANGE
4 NOV 94
PERIODIC
TABLE
SPECint92 94.8 113.3 +20%
SPECfp92 72.0 73.7 + 2%
-- Don Summers
Physics Dept.
University of Mississippi
Oxford, MS 38677
Mike Guttman
>In article <3n2e18$7...@ixnews3.ix.netcom.com>,
>fl...@ix.netcom.com (Flo Rueb) writes:
>>
>>I just have a question regarding the line of MIPS processors. Which of
>>the one: R4400, R4600 or R4700 is faster.
>>
>the 133MHz R4600 out and just plug in the 150MHz R4700 along with a 50MHz clock
>chip to replace the old 44 MHz clock chip.
This is the killer...
I haven't actually looked, but you're saying that the CPU chip is
soldered on to the board. Too bad SGI made it unfeasible for most
of us to perform these simple, relatively inexpensive CPU upgrades.
Though, at least they gave us slots for adding RAM. :-)
BTW, thanks for those tables comparing the CPUs!
- Mike
Martin Knoblauch
just to make it clear: the CPU is not located on the Indy-"mother-
board", but on a user-replaceable daughter card. On that card
(I looked at a spare part, don't do this at home kids :-) the
CPU chip is socketed (at least the one I looked at). So in theory
(voiding your warranty by just thinking of it :-) you could
replace the CPU if you find a 100% pin and function compatible
one. But I would not want to make a bet whether a R4700 would
boot up a current version of IRIX (or even the PROM monitor),
even if you manage not to break the CPU module.