smiller@wet.UUCP (Gregory Shane Miller) (09/22/90)
21 September 1990
I too have been waiting for the release of NeXT's new
computer. But I'm wondering what the "big picture" is.
Byte's November 1988 issue (review on original cube) quotes
Jobs writing: "'MIPS is only one-third of the equation;
sustained system throughput is the key'".
This is absolutely correct.
No kidding! Look at *all* hardware NeXT added to keep work
off the CPU:
NCR 53C90 SCSI Chip [ moves 4 MB / sec ]
Motorola 20 MHz DSP 56001
2 custom VSLI chips for handling
SCSI interface
optical drive
serial ports (2)
ethernet port
12 DMA channels
NeXT uses the Mach Unix kernel (from Carnegie-Mellon). The
September 1989 issue of Unix World writes: "Last, but not
least, Next uses Mach. Mach is a new operating system kernel
that can run 25 percent faster than a UNIX kernel".
So you can imagine how depressed and (frankly) confused I was
when in the March 1990 issue of Byte I find the new UNIX
benchmarks which includes times from a (030) NeXT machine:
Everex i386/33 Mhz SCO Xenix 2.3.1: Cum. Index 6.00
NeXT : Cum. Index 3.54
DEC3100 : Cum. Index 11.12
HP370 : Cum. Index 5.37
The baseline was the Everex; the NeXT ran half as fast as the
baseline, HP370 5/6 ths as fast, DEC3100 nearly twice as
fast as the baseline.
Note:
[1] The DEC3100 may be (I don't know) many more times
expensive than a NeXT and so it *should* run fast.
[2] The article did not give the RAM size, OS version,
or the like for the NeXT, DEC3100, or HP370.
Why is the 030 NeXT box so slow? If the benchmark tests were
unfair, how so? Unfair enough so that NeXT was half as slow
as a i386/33Mhz box?
My point is this: If NeXT just took out the 68030/68882 chips
and replaced them with their 040 relatives, where is the
improvement in the throughput?
I am forced to admit that the new price ($4995) with options
of color boards is exactly what the doctor ordered.
However, for multitasking or multi-user, the net result may
likely be the same as the old cube. On the other hand,
Walter C. Daugherity (usenet@helios.TAMU.EDU) writes
(19 Sept 1990) concerning the new NeXT box:
[ text deleted ect. ]
>NeXT $333 /MIP for 15 MIPS (list price on system $4995)
>Sun Sparc SLC $500 /MIP for 13 MIPS (list price on system $6500)
>386/Mac ci $2000/MIP for 1.75 MIPS (list not given [approx $3500?])
Thus, I am forced to ask three questions:
[1] Does the new NeXT have more "slave" chips? Faster
versions of the old "slave" chips?
[2] Given a 68030 and a 68040 of same speed, is the 040
really faster? If so, by how much? Does the 040
have more RISC architecture? More parallel instruction
decoding? More pipelining? [Assembler programmers,
what's the big picture here?]
[3] Apple has really done a nice job with the MACs and networking.
Universities (et al) just how simple is it to setup (eg. hard-
ware and software) a NeXT net? How's the throughput?
BTW: September 1990 Unix World has this to say:
"At this point, I see administering to a NeXT - or a
network of NeXTs - as a big problem for non-technical
users."
[ again, on the other hand ] earlier in the same article:
"I was pleased to discover that NeXT had not made the
same mistake that Sun did with the Sun 386 - assume
that the network would have nothing but identical
computers on it."
What about gentlemen? Does the 040 box have the same throughput?
Regards -
Shane Millerrca@cs.brown.edu (Ronald C.F. Antony) (09/25/90)
In article <1558@wet.UUCP> smiller@wet.UUCP (Gregory Shane Miller) writes: > Byte's November 1988 issue (review on original cube) quotes > Jobs writing: "'MIPS is only one-third of the equation; > sustained system throughput is the key'". This is when byte still was a bit more usefull than it is today > So you can imagine how depressed and (frankly) confused I was > when in the March 1990 issue of Byte I find the new UNIX > benchmarks which includes times from a (030) NeXT machine: > > Everex i386/33 Mhz SCO Xenix 2.3.1: Cum. Index 6.00 > NeXT : Cum. Index 3.54 > DEC3100 : Cum. Index 11.12 > HP370 : Cum. Index 5.37 The problem with this tests are many. As far as I know they used the base version of a NeXT for these tests. This means only 8 MByte Ram. The NeXT becomes about twice as fast with more memory. After all a lot of the memory is used for the window system. 2nd they tested UNIX features what ever this means. So they tried to compare a UNIX-PC that is supposed to run a couple of terminals in text mode with a graphics oriented workstations called NeXT and another one called DEC3100. Only that the DEC didn't use postscript. Now immagine their nice tests producing output: one blasts it on a 80x25 character display, another one uses X and the NeXT reders the output in PstScript. What a fair contest... Guess how much was lost there! If they has tried the same benchmarks with standard terminals hooked up to each of the machines, no X or NeXTStep running, then we might have gotten results that come closer to reality. But, what do you expect from byte? It's only a shame that UnixWorld had a lot of bad errors in their review, too. > [2] Given a 68030 and a 68040 of same speed, is the 040 > really faster? If so, by how much? Does the 040 According to Motorola the 68040 an average instruction in 1.3 clock cycles. This means about 20 MIPS at 25 MHz clock. The 68040 is about 2-4 times faster than the 68030 on integer ops. and about 3-10 faster than the 68882 on floating point ops. > BTW: September 1990 Unix World has this to say: > > "At this point, I see administering to a NeXT - or a > network of NeXTs - as a big problem for non-technical > users." This is one of the bad errors in this review. They didn't mention the NetInfoManager, nor did they realize that with niload and nidump it is possible to use all the conventional shell scripts for admin. purposes. But then, administering of a UNIX-network never was for non-tecnical users. From time to time you have to look after certain files... But NeXT did a good job anyway. Now with the NetInfoKit it is probably even easier to create administarion Tools. > What about gentlemen? Does the 040 box have the same throughput? The 'old' NeXT wasn't as bad a people said, but I would be really surprised if the new ones weren't still a lot better Ronald ------------------------------------------------------------------------------ "The reasonable man adapts himself to the world; the unreasonable one persists in trying to adapt the world to himself. Therefore all progress depends on the unreasonable man." Bernhard Shaw | rca@cs.brown.edu or antony@browncog.bitnet
edwardj@microsoft.UUCP (Edward JUNG) (09/28/90)
In article <50939@brunix.UUCP> rca@cs.brown.edu (Ronald C.F. Antony) writes: [stuff deleted] >2nd they tested UNIX features what ever this means. So they tried to >compare a UNIX-PC that is supposed to run a couple of terminals in >text mode with a graphics oriented workstations called NeXT and >another one called DEC3100. Only that the DEC didn't use postscript. >Now immagine their nice tests producing output: one blasts it on a >80x25 character display, another one uses X and the NeXT reders the >output in PstScript. What a fair contest... [more deleted] I don't want to get into a benchmark war. However, if the Dec3100 wasn't running display postscript, then what was it running? Was BYTE running the DecStation after disabling DPS in some manner? If so, what window manager and display library *were* they running? -- Edward Jung Microsoft Corp. My opinions do not reflect any policy of my employer.