Unpopular opinion, NeXT hardware was kinda junk. It was a 68030 machine released into direct competition SPARC and MIPS chips that were 3x faster. It had a 2bpp framebuffer (and UI) that would have been really clever and attractive in competition with monochrome workstations of 1986, but frankly when it shipped it looked kinda awful next to the 1MB SVGA cards arriving on the market. At the OS level, Mach was outrageously primitive for a commercial unix (no shared libraries!). And the one genuinely unique and innovative feature (the big MO drive) turned out to be a technological dead end.
The NeXTStep layer was really the only selling point. And in hindsight that was enough. But the hardware itself? Only barely worth remembering.
The original OS absolutely had shared libraries - the whole environment was so RAM-constrained that it was essential. (source - I was part of the early SW eng team). In fact, my memory is that they worked much better than competing systems because of Objective-C's late binding, in the practical case of sharing library code across applications built/shipped at different times on different OS versions. As I recall, C++ based systems couldn't deliver the full sharing on a typical user's system who was never running apps that were all updated for each major and minor release.
I was in College at the time, comparing it to the Dec and VaxStations next to it, it held it's own. 'the one genuinely unique feature'?
So Display postscript and WYSIWYG _everywhere_ wasn't a big deal? The developer environment and DSP capabilities were pretty special....the Magnesium was a cool flex and one cable going to the monitor keyboard and mouse...the MO drive was slow, but cool.
> I was in College at the time, comparing it to the Dec and VaxStations next to it, it held it's own
It's interesting to me that this criticism has stayed around so many years in different forms: the iPhone has underpowered RAM compared to Androids, etc etc.
Yet consistently people pick on the overall experience not particular components.
IIRC it was the fastest workstation available for under $5k in 1990 (around 15 mips, I think?). You have to remember what market they thought they were selling to -- it was meant to be a high-end personal workstation, not an industrial machine. Sun and SGI machines started at around $10k and went up from there.
They were sold, among other places, in college bookstores, right beside the Mac and Windows machines. I was a campus consultant and sold a number of them - a lot of my job was hanging out in the bookstore and talking to potential buyers. I have no idea how one went about buying a Sun, but they certainly didn't bother much with individual sales.
So it's competition was the higher end of the Mac line and a bit up from that, really. Turned out there was never much of a market for personal workstations, hence the low sales numbers.
I came into a bit of money in college and decided to buy a Sun workstation.
You are correct - they didn’t know how to fulfill individual sales. The box was delivered to a LTL freight crossdock, and I got a call from a dispatcher to come pick the thing up. I had to drive to the industrial part of town. A bemused guy helped me get the carton down and remove the boxes, then I got to lug thing across the yard.
It was a fun thing to have amongst my nerd friends. I ended up selling it at a nice profit because I was willing to ship it anywhere. The monitor i sold at a loss to a graduate student.
I had another boss at the same company that filled a NeXT cube to capacity with motherboards. For the sake of a good story I will say he had them all running under the same Mach kernel. He described the simple thing he did to get it running, but I have forgotten.
The additional boards were netbooted via bootp/tftp, so the additional boards could run the same kernel as the one the hard disk (and/or MO drive) was connected to. Of course, this is a distributed memory machine - not much different than running three diskless Cubes booting from a fourth machine.
The NeXT was priced above a PC, but just below a SGI, DEC or Sun workstation. When compared to a top of the line PS/2 model 80 (at $4500 for the tower, and $1000 for the monitor, plus an extra $200 for Windows 386), the NeXT Cube ($6K or so) was a pretty good value because you got a laser printer, the FPU was included and the software library that came with it came with development tools (everyone else charged for them), FrameMaker, Mathmatica and a bunch of other titles that would have been $450 each on the PS/2.
PS/2 Model 80 release MSRPs were $6,995, $8,295, and $10,995, depending on the configuration, without monitor, OS, FPU, or support for display resolutions beyond standard VGA[1]. The top model included a 20 MHz CPU, 2 MB RAM, and a 115 MB ESDI hard drive.
NeXT shipped two years later, but its $9,995 MSRP[2] included the OS, application software, additional software, a 17" 1120x832 greyscale monitor, FPU, DSP, and four times the RAM of the top-end Model 80.
IIRC — and I may not, it's been decades — NeXT generally sold at or around MSRP, while other UNIX vendors and IBM typically sold at a considerable discount to large accounts and academia.
To compare with a more "garden-variety" PC, in the same BYTE issue[2] as the NeXT article, Dell was selling its lowest-end 386 for $4,199, with a 25 MHz CPU, 1 MB RAM, monochrome VGA display (640x480, size unspecified), and 40 MB (interface unspecified, probably IDE) hard drive for $6,299, no FPU, OS extra.
This same Dell system with a 150 MB hard drive, SVGA color display (800x600, size unspecified), and 8 MB RAM cost around $7,000 ($5,900 for 150 MB SVGA, +$600 for 4 MB RAM; 8 MB, I presume, would cost another $600×4/(4-1) = $800 or so).
Adding a $500 387/25 FPU (based on ads in the back of the same BYTE[2]) and a $500 Dell UNIX license[3] brings the cost of our Dell system to $8,000, so within 20% or so of NeXT.
Unlike Dell, NeXT also bundled a variety of high-quality developer tools (Allegro Common Lisp and Sybase SQL Server in addition to the expected GNU C/ObjC toolchain, Interface Builder, and NS class libraries), useful application software (Mathematica, WriteNow, Mail.app), and convenient reference works (dictionary, thesaurus, book of quotations, complete works of Shakespeare, product docs).
Finally, NeXT charged $3,695 to upgrade the Cube's internal hard drive from 40 MB to 330 MB, which makes Apple SSD upgrade pricing sound entirely reasonable, as a comparable upgrade from a hypothetical 512 GB to 4 TB NeXT SSD would cost almost $3,000.
i don't know about the CPU, but the production quality of the hardware was great. once we had a microsecond poweroutage, the PC powered off, the NeXT machine kept running.
> The 68030 does look like a curious choice for a machine launched in 1988.
GP is exaggerating. Spark and MIPS were more expensive machines for a different market NeXT wanted a toe in but was merely straddling; business and education were NeXT's target markets. The 68030 with a math coprocessor was bleeding edge in 1988. Sun released a 20MHz and 33MHz 68030 Sun-3x in 1989, a SunOS (UNIX) workstation.[1] Apple released the 68030 Mac SE/30 (16MHz) in Jan. 1989. There was absolutely nothing wrong with the 68030 in 1988/1989. If anything, NeXT was a little fast out of the gate with the 25MHz 68030 in 1988.
They developed a dual-CPU prototype box based on the 88000, circa 1992. It was never released - the company was bought before it advanced past the prototype stage.
They also had a prototype laptop, although I don't know for certain which CPU it was based on. Because the existing battery technology couldn't provide enough juice for reasonable runtime, it was going to be a plug-into-the-wall design.
They gave up on the 88000 the same time Apple did (yes it was considered at Apple too) and is directly related the IBM/Apple PPC partnership and Apple’s insistence that Motorola being a second source for the PPC. It was really Motorola that gave up on the 88000 and not Next.
NeXT wasn't bought until late 1996. That didnt kill the RISC project, what happened is the company dropped all hardware and became software-only (primarily on top of x86 boxes, but some lower-level pieces ran on HP and SPARC).
As part of a project code-named "Jaguar", Apple did consider using a Motorola 88K variant as their future RISC-based hardware, but it was short-lived and they moved on to the POWER Architecture. In 1991, Apple, IBM, and Motorola formed the "AIM" alliance with the goal of creating a Common Hardware Reference Platform (CHRP). This led to the PowerPC architecture, which included most of the POWER instructions along with some additions and deletions.
The NeXTStep layer was really the only selling point. And in hindsight that was enough. But the hardware itself? Only barely worth remembering.