Graphics from Intel is much more believable, probably something that later became Intel i860 based NeXTdimension. But that would push whole story a couple years since Intel i860 released mid 1989.
That was a RISC CPU many times faster than MC68030+MC68882 (it had a pipelined FPU, like Intel 80860 and the much later Intel Pentium, so for number crunching it was more than an order of magnitude faster than MC68030+MC68882).
It probably was used as an accelerator, on an expansion card. NeXT had NuBus for expansion cards. I do not know if it had enough connectors for the 2 expansion cards wanted by CIA, but if any customization was done it must have consisted in providing enough NuBus connectors.
The Intel graphics could have been Intel 82720, a clone of NEC μPD7220, which could do by itself simple 2D graphics tasks (like drawing lines and arcs, filling polygons and copying the font bitmaps for rendering character strings), freeing the main CPU.
By the end of 1990, my company bought what at that time were expensive workstation variants of an IBM PC/AT clone, which had some huge monitors with higher resolution than supported by VGA, for which an expansion card with Hitachi HD63484 was used to provide really impressive graphics.
Intel 82720 was similar to Hitachi HD63484, but HD63484 was more recent and it had more features, e.g. it supported up to 4k resolution (but 4k monitors were extremely rare and expensive in 1990, so we had only monitors with 768x1024 @ 16 bit per pixel, which was still excellent for that time).
So I assume that a similar card was referred to in the article.
We used our HD63484 card only with AutoCAD, because it was the only program that had graphics drivers for it. The rest of the programs used a standard VGA card and a VGA monitor.
I assume that a similar dual monitor arrangement must have been used by CIA/NSA with NeXT, so their special application program must have used their graphics card with a big monitor, while the standard NeXT programs must have used the original NeXT graphics.
So all Next had to do was
1 Develop another set of ASICs for fully incompatible CPU bus in parallel and in secret
2 Develop another motherboard in parallel and in secret
3 Manufacture another motherboard in parallel and in secret
4 Port NextStep to another fully incompatible CPU arch in parallel and in secret
5 Kill all the developers involved so none of this ever came out
> accelerator, on an expansion card
graphics on expansion, accelerator on expansion, additional HDD, new color monitor, new OS. Sound like in your alternate reality NSA paid $6K a pop for 20000 magnesium boxes with new incompatible with anything NuBus backplane offering whole 3 free slots, so $2K per slot.
The interface bus of MC88100 was actually the same that was inherited by the first PowerPC CPUs a few years later, when Apple, IBM and Motorola cooperated to make the PowerPC specification, based on the IBM POWER ISA.
It would not have needed any ASICs, but just some PALs to interface with the on-card memory. The expansion card would only have needed to make the on-card memory also readable and writable by the host MC68030 CPU through NuBus, which would also have required only some PALs, no ASICs.
Your price estimate is likely to be around the correct value, so there is no doubt that CIA/NSA did not try to save pennies when buying all this hardware.
The NeXT computer was more expensive than an IBM PC/AT, but otherwise in all this configuration there is only the MC88100 card that is added in comparison with the CAD workstations that I have used myself in 1990, at a company with a moderate budget for such acquisitions, so there is no doubt that US TLA agencies could purchase something better.
A MC88100 accelerator card would have cost as much as an entire computer with its peripherals, but this just means that such a workstation would have cost as much as 2 standard CAD workstations, like the ones I used. Surely CIA/NSA could afford that.
"The physical pinout of the DSP56001 is different from the DSP56002" https://digisolutions.de/downloads/Differences56001_56002.pd...
Buying new unproven workstation just to develop expansion cards to make it fast makes no sense when you can just buy off teh shelf fast SUN or SGI.
The only possibility is April 1989 announcement of 50 MHz versions of 68030/68882 and Clock Doubler circuit running CPU/FPU fast only when its not touching the bus. Afaik to this day this is the only way to OC Nexts.
https://68kmla.org/bb/threads/nextstation-clock-doubler.4979...
https://web.archive.org/web/20190722120829/https://www.nextc...
But this still pushes whole thing to ~1990. Would NSA get 20000 highly binned CPUs before actual volume production? Then there is absolute lack of evidence of even one such system ever surviving, and stuff does survive despite being classified, like Tempest IBM 5170s.
Motorola already had MC88100+MC88200, a RISC CPU with pipelined FPU, which was many times faster than MC68030+MC68882. For floating-point computations it might have been at least 20 times faster than MC68030+MC68882.
NeXT had NuBus for expansion cards, so it would not have been any problem to have a NuBus accelerator card with MC88100+MC88200 and a decent amount of memory for that time, e.g. 16 MB or 32 MB.
MC88100 had been developed in parallel with MC68030. I remember very well the advertisements for MC88100 from 1988, because for some time it was the fastest monolithic CPU (until Intel launched the similar 80860 a year later, which had a higher clock frequency, because of being made with a newer CMOS fabrication process, the same which was used for the first 80486). The advertisements were that now you could have your own supercomputer by choosing MC88100, instead of paying millions for a Cray.
Wikipedia even says that Jobs wanted in the beginning to design NeXT with MC88100, but he switched to MC68030 due to price and availability problems, which should not have been a problem for the CIA.